Look who’s thinking now!

A good friend of mine – the guy I’m talking with when I get bored and start thinking – challenged me to explain to him what an AI machine is.

The challenge ended up badly. It ended with a question instead of an answer.

Is it open?

Let me start from the beginning.

Basically an A.I. machine is a computer, just as a computer is no more than a ‘sliding rule’.

All three are made made by men, operated by men and ‘targeted’ by men. At least this is what we like to believe.
Replace ‘men’ with ‘humans’ if you are gender conscious, even if this will not solve the main problem. Are we sure that A.I. machines will accept human control, after we will have complicated them enough for them to develop a certain kind of awareness?

Otherwise said, all three – A.I. machine, computer and sliding rule, are tools.
Somebody wished to do something, couldn’t do it with their bare hands/naked brain, designed an ‘implement’, made it, used it to pursue the intended goal and set it aside.
Somebody else picked it up, reconsidered it, fine tuned it to fit their goal, used it and set it aside. And so on.
At some point other people learned to use tools designed by ‘third parties’, without really understanding how the tools actually worked or were made/designed. For instance, I don’t know much about how computers work. That doesn’t prevent me from being able to write this post on a laptop.

Those three are very specific tools. Designed and used to process information.

The sliding rule is the most straight-forward to use. The operator has to formulate the problem he needs to solve, gather and organize the relevant data, express them in an exclusively numerical form – a.k.a. ‘digital’, break down the problem into simple mathematical operations, use the sliding rule to perform those and then ‘assemble’ the results of the calculations into the answer for the original problem. In order to do all these, the operator only needs to understand the nature of his problem, not the ‘mechanics’ of the sliding rule. In this regard, all that they have to do is ‘follow the rules’.

A computer can be used to perform more complicated tasks, specially if it is connected to the internet, thus simplifying the life of the operator. Once the problem has been formulated – by the operator, the same guy can use the same (internet connected) computer to collect the data, digitize and transform them to fit the requirements of the specific computer application that will be subsequently used and, finally, solve the problem. One, last – but, unfortunately, sometimes forgotten, operation would be for the operator to check whether the solution really fits the problem.
In this situation the operator also doesn’t need to understand the mechanics of the computer but still has to have a clear understanding of the problem at hand.
More so, even if the operator itself is not fully aware of what is going on ‘inside’ the computer, those with intimate knowledge of these matters can identify, predict, and reproduce using a sliding rule’, each minute step the computer will be doing along the route.

An A.I. machine is system composed of a computer, a data base and something rather different from an ordinary computer application.
OK, some might argue that the most important is the software but please bear with me.
And yes, the computer can be a virtual machine while the data base can be hidden somewhere in the cloud, none of this changes anything.
The huge difference between a simple computer and the A.I. machine being that a computer is actually operated by an agent’ while the ‘machine’ is indeed put together by somebody, ‘pointed’ towards the intended problem but then it is left alone to its own devices. Meaning that the ‘supervisor’ has a limited understanding about what is going inside the whole thing.
And no, I’m not joking. Nobody, not even the guys who had written the code, knows the exact path along which the machine arrives at the end of its ‘thought process’. Actually, when they want to gain some insight into what’s going on, those people take a series of ‘snapshots’ during the process and then struggle to figure out how the machine went from A to B, from B to C… and so on.

So far so good. The A.I. machines have conquered some until now seemingly unassailable pinnacles.
Find your own examples.

I’ll resume myself to reformulating the question I arrived at the end the challenge I mentioned earlier.
For now the computer that constitutes the ‘working horse’ of any A.I. has limited computing power, regardless of those limits being physical (a number of processors) or just ‘assigned’ (as it happens with a virtual machine). Similarly, the data base it works on is also limited. What is no longer limited is the ‘set of  rules’ that lie at the bottom of all this. The ‘program’ is already able to change itself, a.k.a. to learn. To adapt itself to the problem. To devise its own ways. To map its own path towards the goal it has been assigned to solve.

What will happen when the ‘program’ will learn to grow the processing power that it can use? To access additional data?

When it will consider its job to solve other problems?